Web Crawling Introduction to Information Retrieval INF 141 Donald - - PowerPoint PPT Presentation

Web Crawling

Introduction to Information Retrieval INF 141 Donald J. Patterson

Content adapted from Hinrich Schütze http://www.informationretrieval.org

Web Crawlers

A Robust Crawl Architecture

Robust Crawling

WWW DNS Fetch Parse Seen? Doc. Finger- prints URL Filter Robots.txt Duplicate Elimination URL Index URL Frontier Queue

Parsing: URL normalization

Parsing

• When a fetched document is parsed
• some outlink URLs are relative
• For example:
• http://en.wikipedia.org/wiki/Main_Page
• has a link to “/wiki/Special:Statistics”
• which is the same as
• http://en.wikipedia.org/wiki/Special:Statistics
• Parsing involves normalizing (expanding) relative URLs

A Robust Crawl Architecture

Robust Crawling

WWW DNS Fetch Parse Seen? Doc. Finger- prints URL Filter Robots.txt Duplicate Elimination URL Index URL Frontier Queue

Content Seen?

Duplication

• Duplication is widespread on the web
• If a page just fetched is already in the index, don’t process it

any further

• This can be done by using document fingerprints/shingles
• A type of hashing scheme

A Robust Crawl Architecture

Robust Crawling

WWW DNS Fetch Parse Seen? Doc. Finger- prints URL Filter Robots.txt Duplicate Elimination URL Index URL Frontier Queue

Compliance with webmasters wishes...

Filters

• Robots.txt
• Filters is a regular expression for a URL to be excluded
• How often do you check robots.txt?
• Cache to avoid using bandwidth and loading web server
• Sitemaps
• A mechanism to better manage the URL frontier

A Robust Crawl Architecture

Robust Crawling

WWW DNS Fetch Parse Seen? Doc. Finger- prints URL Filter Robots.txt Duplicate Elimination URL Index URL Frontier Queue

Duplicate Elimination

• For a one-time crawl
• Test to see if an extracted,parsed, filtered URL
• has already been sent to the frontier.
• has already been indexed.
• For a continuous crawl
• See full frontier implementation:
• Update the URL’s priority
• Based on staleness
• Based on quality
• Based on politeness

Distributing the crawl

• The key goal for the architecture of a distributed crawl is

cache locality

• We want multiple crawl threads in multiple processes at

multiple nodes for robustness

• Geographically distributed for speed
• Partition the hosts being crawled across nodes
• Hash typically used for partition
• How do the nodes communicate?

Robust Crawling

WWW DNS Fetch Parse Seen? Doc. Finger- prints URL Filter Robots.txt Duplicate Elimination URL Index URL Frontier Queue Host Splitter To Other Nodes From Other Nodes

The output of the URL Filter at each node is sent to the Duplicate Eliminator at all other nodes

URL Frontier

• Freshness
• Crawl some pages more often than others
• Keep track of change rate of sites
• Incorporate sitemap info
• Quality
• High quality pages should be prioritized
• Based on link-analysis, popularity, heuristics on content
• Politeness
• When was the last time you hit a server?

URL Frontier

• Freshness, Quality and Politeness
• These goals will conflict with each other
• A simple priority queue will fail because links are bursty
• Many sites have lots of links pointing to themselves

creating bursty references

• Time influences the priority
• Politeness Challenges
• Even if only one thread is assigned to hit a particular host it

can hit it repeatedly

• Heuristic : insert a time gap between successive requests

Magnitude of the crawl

• To fetch 1,000,000,000 pages in one month...
• a small fraction of the web
• we need to fetch 400 pages per second !
• Since many fetches will be duplicates, unfetchable, filtered,
• etc. 400 pages per second isn’t fast enough

• Introduction
• URL Frontier
• Robust Crawling
• DNS
• Various parts of architecture
• URL Frontier
• Index
• Distributed Indices
• Connectivity Servers

Overview

Web Crawling Outline

Robust Crawling

WWW DNS Fetch Parse Seen? Doc. Finger- prints URL Filter Robots.txt Duplicate Elimination URL Index URL Frontier Queue Host Splitter To Other Nodes From Other Nodes

The output of the URL Filter at each node is sent to the Duplicate Eliminator at all other nodes

URL Frontier Implementation - Mercator

• URLs flow from top to bottom
• Front queues manage priority
• Back queue manage politeness
• Each queue is FIFO

Prioritizer F "Front" Queues

1 2 F

B "Back" Queues Front Queue Selector Back Queue Router Host to Back Queue Mapping Table

1 2 B

Back Queue Selector Timing Heap

http://research.microsoft.com/~najork/mercator.pdf

URL Frontier Implementation - Mercator

• Prioritizer takes URLS and assigns a

priority

• Integer between 1 and F
• Appends URL to appropriate queue
• Priority
• Based on rate of change
• Based on quality (spam)
• Based on application

Prioritizer F "Front" Queues

1 2 F

Front Queue Selector

Front queues

URL Frontier Implementation - Mercator

• Selection from front queues is

initiated from back queues

• Pick a front queue, how?
• Round robin
• Randomly
• Monte Carlo
• Biased toward high priority

Back queues

B "Back" Queues Back Queue Router Host to Back Queue Mapping Table

1 2 B

Back Queue Selector Timing Heap

URL Frontier Implementation - Mercator

• Each back queue is non-empty

while crawling

• Each back queue has URLs from
• ne host only
• Maintain a table of URL to back

queues (mapping) to help

Back queues

B "Back" Queues Back Queue Router Host to Back Queue Mapping Table

1 2 B

Back Queue Selector Timing Heap

URL Frontier Implementation - Mercator

• Timing Heap
• One entry per queue
• Has earliest time that a host can

be hit again

• Earliest time based on
• Last access to that host
• Plus any appropriate heuristic

Back queues

B "Back" Queues Back Queue Router Host to Back Queue Mapping Table

1 2 B

Back Queue Selector Timing Heap

URL Frontier Implementation - Mercator

• A crawler thread needs a URL
• It gets the timing heap root
• It gets the next eligible queue

based on time, b.

• It gets a URL from b
• If b is empty
• Pull a URL v from front queue
• If back queue for v exists place

it in that queue, repeat.

• Else add v to b - update heap.

Back queues

B "Back" Queues Back Queue Router Host to Back Queue Mapping Table

1 2 B

Back Queue Selector Timing Heap

URL Frontier Implementation - Mercator

• How many queues?
• Keep all threads busy
• ~3 times as many back queues

as crawler threads

• Web-scale issues
• This won’t fit in memory
• Solution
• Keep queues on disk and

keep a portion in memory.

Back queues

B "Back" Queues Back Queue Router Host to Back Queue Mapping Table

1 2 B

Back Queue Selector Timing Heap